fs/xfs/xfs_dir2_block.c

   1 /*
   2  * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
   3  * Copyright (c) 2013 Red Hat, Inc.
   4  * All Rights Reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License as
   8  * published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it would be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program; if not, write the Free Software Foundation,
  17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  18  */
  19 #include "xfs.h"
  20 #include "xfs_fs.h"
  21 #include "xfs_types.h"
  22 #include "xfs_log.h"
  23 #include "xfs_trans.h"
  24 #include "xfs_sb.h"
  25 #include "xfs_ag.h"
  26 #include "xfs_mount.h"
  27 #include "xfs_da_btree.h"
  28 #include "xfs_bmap_btree.h"
  29 #include "xfs_dinode.h"
  30 #include "xfs_inode.h"
  31 #include "xfs_inode_item.h"
  32 #include "xfs_bmap.h"
  33 #include "xfs_buf_item.h"
  34 #include "xfs_dir2.h"
  35 #include "xfs_dir2_format.h"
  36 #include "xfs_dir2_priv.h"
  37 #include "xfs_error.h"
  38 #include "xfs_trace.h"
  39 #include "xfs_cksum.h"
  40
  41 /*
  42  * Local function prototypes.
  43  */
  44 static void xfs_dir2_block_log_leaf(xfs_trans_t *tp, struct xfs_buf *bp,
  45                                     int first, int last);
  46 static void xfs_dir2_block_log_tail(xfs_trans_t *tp, struct xfs_buf *bp);
  47 static int xfs_dir2_block_lookup_int(xfs_da_args_t *args, struct xfs_buf **bpp,
  48                                      int *entno);
  49 static int xfs_dir2_block_sort(const void *a, const void *b);
  50
  51 static xfs_dahash_t xfs_dir_hash_dot, xfs_dir_hash_dotdot;
  52
  53 /*
  54  * One-time startup routine called from xfs_init().
  55  */
  56 void
  57 xfs_dir_startup(void)
  58 {
  59         xfs_dir_hash_dot = xfs_da_hashname((unsigned char *)".", 1);
  60         xfs_dir_hash_dotdot = xfs_da_hashname((unsigned char *)"..", 2);
  61 }
  62
  63 static bool
  64 xfs_dir3_block_verify(
  65         struct xfs_buf          *bp)
  66 {
  67         struct xfs_mount        *mp = bp->b_target->bt_mount;
  68         struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
  69
  70         if (xfs_sb_version_hascrc(&mp->m_sb)) {
  71                 if (hdr3->magic != cpu_to_be32(XFS_DIR3_BLOCK_MAGIC))
  72                         return false;
  73                 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_uuid))
  74                         return false;
  75                 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
  76                         return false;
  77         } else {
  78                 if (hdr3->magic != cpu_to_be32(XFS_DIR2_BLOCK_MAGIC))
  79                         return false;
  80         }
  81         if (__xfs_dir3_data_check(NULL, bp))
  82                 return false;
  83         return true;
  84 }
  85
  86 static void
  87 xfs_dir3_block_read_verify(
  88         struct xfs_buf  *bp)
  89 {
  90         struct xfs_mount        *mp = bp->b_target->bt_mount;
  91
  92         if ((xfs_sb_version_hascrc(&mp->m_sb) &&
  93              !xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
  94                                           XFS_DIR3_DATA_CRC_OFF)) ||
  95             !xfs_dir3_block_verify(bp)) {
  96                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
  97                 xfs_buf_ioerror(bp, EFSCORRUPTED);
  98         }
  99 }
 100
 101 static void
 102 xfs_dir3_block_write_verify(
 103         struct xfs_buf  *bp)
 104 {
 105         struct xfs_mount        *mp = bp->b_target->bt_mount;
 106         struct xfs_buf_log_item *bip = bp->b_fspriv;
 107         struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 108
 109         if (!xfs_dir3_block_verify(bp)) {
 110                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
 111                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 112                 return;
 113         }
 114
 115         if (!xfs_sb_version_hascrc(&mp->m_sb))
 116                 return;
 117
 118         if (bip)
 119                 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
 120
 121         xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length), XFS_DIR3_DATA_CRC_OFF);
 122 }
 123
 124 const struct xfs_buf_ops xfs_dir3_block_buf_ops = {
 125         .verify_read = xfs_dir3_block_read_verify,
 126         .verify_write = xfs_dir3_block_write_verify,
 127 };
 128
 129 static int
 130 xfs_dir3_block_read(
 131         struct xfs_trans        *tp,
 132         struct xfs_inode        *dp,
 133         struct xfs_buf          **bpp)
 134 {
 135         struct xfs_mount        *mp = dp->i_mount;
 136         int                     err;
 137
 138         err = xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, bpp,
 139                                 XFS_DATA_FORK, &xfs_dir3_block_buf_ops);
 140         if (!err && tp)
 141                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_BLOCK_BUF);
 142         return err;
 143 }
 144
 145 static void
 146 xfs_dir3_block_init(
 147         struct xfs_mount        *mp,
 148         struct xfs_trans        *tp,
 149         struct xfs_buf          *bp,
 150         struct xfs_inode        *dp)
 151 {
 152         struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 153
 154         bp->b_ops = &xfs_dir3_block_buf_ops;
 155         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_BLOCK_BUF);
 156
 157         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 158                 memset(hdr3, 0, sizeof(*hdr3));
 159                 hdr3->magic = cpu_to_be32(XFS_DIR3_BLOCK_MAGIC);
 160                 hdr3->blkno = cpu_to_be64(bp->b_bn);
 161                 hdr3->owner = cpu_to_be64(dp->i_ino);
 162                 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_uuid);
 163                 return;
 164
 165         }
 166         hdr3->magic = cpu_to_be32(XFS_DIR2_BLOCK_MAGIC);
 167 }
 168
 169 static void
 170 xfs_dir2_block_need_space(
 171         struct xfs_dir2_data_hdr        *hdr,
 172         struct xfs_dir2_block_tail      *btp,
 173         struct xfs_dir2_leaf_entry      *blp,
 174         __be16                          **tagpp,
 175         struct xfs_dir2_data_unused     **dupp,
 176         struct xfs_dir2_data_unused     **enddupp,
 177         int                             *compact,
 178         int                             len)
 179 {
 180         struct xfs_dir2_data_free       *bf;
 181         __be16                          *tagp = NULL;
 182         struct xfs_dir2_data_unused     *dup = NULL;
 183         struct xfs_dir2_data_unused     *enddup = NULL;
 184
 185         *compact = 0;
 186         bf = xfs_dir3_data_bestfree_p(hdr);
 187
 188         /*
 189          * If there are stale entries we'll use one for the leaf.
 190          */
 191         if (btp->stale) {
 192                 if (be16_to_cpu(bf[0].length) >= len) {
 193                         /*
 194                          * The biggest entry enough to avoid compaction.
 195                          */
 196                         dup = (xfs_dir2_data_unused_t *)
 197                               ((char *)hdr + be16_to_cpu(bf[0].offset));
 198                         goto out;
 199                 }
 200
 201                 /*
 202                  * Will need to compact to make this work.
 203                  * Tag just before the first leaf entry.
 204                  */
 205                 *compact = 1;
 206                 tagp = (__be16 *)blp - 1;
 207
 208                 /* Data object just before the first leaf entry.  */
 209                 dup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
 210
 211                 /*
 212                  * If it's not free then the data will go where the
 213                  * leaf data starts now, if it works at all.
 214                  */
 215                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 216                         if (be16_to_cpu(dup->length) + (be32_to_cpu(btp->stale) - 1) *
 217                             (uint)sizeof(*blp) < len)
 218                                 dup = NULL;
 219                 } else if ((be32_to_cpu(btp->stale) - 1) * (uint)sizeof(*blp) < len)
 220                         dup = NULL;
 221                 else
 222                         dup = (xfs_dir2_data_unused_t *)blp;
 223                 goto out;
 224         }
 225
 226         /*
 227          * no stale entries, so just use free space.
 228          * Tag just before the first leaf entry.
 229          */
 230         tagp = (__be16 *)blp - 1;
 231
 232         /* Data object just before the first leaf entry.  */
 233         enddup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
 234
 235         /*
 236          * If it's not free then can't do this add without cleaning up:
 237          * the space before the first leaf entry needs to be free so it
 238          * can be expanded to hold the pointer to the new entry.
 239          */
 240         if (be16_to_cpu(enddup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 241                 /*
 242                  * Check out the biggest freespace and see if it's the same one.
 243                  */
 244                 dup = (xfs_dir2_data_unused_t *)
 245                       ((char *)hdr + be16_to_cpu(bf[0].offset));
 246                 if (dup != enddup) {
 247                         /*
 248                          * Not the same free entry, just check its length.
 249                          */
 250                         if (be16_to_cpu(dup->length) < len)
 251                                 dup = NULL;
 252                         goto out;
 253                 }
 254
 255                 /*
 256                  * It is the biggest freespace, can it hold the leaf too?
 257                  */
 258                 if (be16_to_cpu(dup->length) < len + (uint)sizeof(*blp)) {
 259                         /*
 260                          * Yes, use the second-largest entry instead if it works.
 261                          */
 262                         if (be16_to_cpu(bf[1].length) >= len)
 263                                 dup = (xfs_dir2_data_unused_t *)
 264                                       ((char *)hdr + be16_to_cpu(bf[1].offset));
 265                         else
 266                                 dup = NULL;
 267                 }
 268         }
 269 out:
 270         *tagpp = tagp;
 271         *dupp = dup;
 272         *enddupp = enddup;
 273 }
 274
 275 /*
 276  * compact the leaf entries.
 277  * Leave the highest-numbered stale entry stale.
 278  * XXX should be the one closest to mid but mid is not yet computed.
 279  */
 280 static void
 281 xfs_dir2_block_compact(
 282         struct xfs_trans                *tp,
 283         struct xfs_buf                  *bp,
 284         struct xfs_dir2_data_hdr        *hdr,
 285         struct xfs_dir2_block_tail      *btp,
 286         struct xfs_dir2_leaf_entry      *blp,
 287         int                             *needlog,
 288         int                             *lfloghigh,
 289         int                             *lfloglow)
 290 {
 291         int                     fromidx;        /* source leaf index */
 292         int                     toidx;          /* target leaf index */
 293         int                     needscan = 0;
 294         int                     highstale;      /* high stale index */
 295
 296         fromidx = toidx = be32_to_cpu(btp->count) - 1;
 297         highstale = *lfloghigh = -1;
 298         for (; fromidx >= 0; fromidx--) {
 299                 if (blp[fromidx].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
 300                         if (highstale == -1)
 301                                 highstale = toidx;
 302                         else {
 303                                 if (*lfloghigh == -1)
 304                                         *lfloghigh = toidx;
 305                                 continue;
 306                         }
 307                 }
 308                 if (fromidx < toidx)
 309                         blp[toidx] = blp[fromidx];
 310                 toidx--;
 311         }
 312         *lfloglow = toidx + 1 - (be32_to_cpu(btp->stale) - 1);
 313         *lfloghigh -= be32_to_cpu(btp->stale) - 1;
 314         be32_add_cpu(&btp->count, -(be32_to_cpu(btp->stale) - 1));
 315         xfs_dir2_data_make_free(tp, bp,
 316                 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
 317                 (xfs_dir2_data_aoff_t)((be32_to_cpu(btp->stale) - 1) * sizeof(*blp)),
 318                 needlog, &needscan);
 319         blp += be32_to_cpu(btp->stale) - 1;
 320         btp->stale = cpu_to_be32(1);
 321         /*
 322          * If we now need to rebuild the bestfree map, do so.
 323          * This needs to happen before the next call to use_free.
 324          */
 325         if (needscan)
 326                 xfs_dir2_data_freescan(tp->t_mountp, hdr, needlog);
 327 }
 328
 329 /*
 330  * Add an entry to a block directory.
 331  */
 332 int                                             /* error */
 333 xfs_dir2_block_addname(
 334         xfs_da_args_t           *args)          /* directory op arguments */
 335 {
 336         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 337         xfs_dir2_leaf_entry_t   *blp;           /* block leaf entries */
 338         struct xfs_buf          *bp;            /* buffer for block */
 339         xfs_dir2_block_tail_t   *btp;           /* block tail */
 340         int                     compact;        /* need to compact leaf ents */
 341         xfs_dir2_data_entry_t   *dep;           /* block data entry */
 342         xfs_inode_t             *dp;            /* directory inode */
 343         xfs_dir2_data_unused_t  *dup;           /* block unused entry */
 344         int                     error;          /* error return value */
 345         xfs_dir2_data_unused_t  *enddup=NULL;   /* unused at end of data */
 346         xfs_dahash_t            hash;           /* hash value of found entry */
 347         int                     high;           /* high index for binary srch */
 348         int                     highstale;      /* high stale index */
 349         int                     lfloghigh=0;    /* last final leaf to log */
 350         int                     lfloglow=0;     /* first final leaf to log */
 351         int                     len;            /* length of the new entry */
 352         int                     low;            /* low index for binary srch */
 353         int                     lowstale;       /* low stale index */
 354         int                     mid=0;          /* midpoint for binary srch */
 355         xfs_mount_t             *mp;            /* filesystem mount point */
 356         int                     needlog;        /* need to log header */
 357         int                     needscan;       /* need to rescan freespace */
 358         __be16                  *tagp;          /* pointer to tag value */
 359         xfs_trans_t             *tp;            /* transaction structure */
 360
 361         trace_xfs_dir2_block_addname(args);
 362
 363         dp = args->dp;
 364         tp = args->trans;
 365         mp = dp->i_mount;
 366
 367         /* Read the (one and only) directory block into bp. */
 368         error = xfs_dir3_block_read(tp, dp, &bp);
 369         if (error)
 370                 return error;
 371
 372         len = xfs_dir2_data_entsize(args->namelen);
 373
 374         /*
 375          * Set up pointers to parts of the block.
 376          */
 377         hdr = bp->b_addr;
 378         btp = xfs_dir2_block_tail_p(mp, hdr);
 379         blp = xfs_dir2_block_leaf_p(btp);
 380
 381         /*
 382          * Find out if we can reuse stale entries or whether we need extra
 383          * space for entry and new leaf.
 384          */
 385         xfs_dir2_block_need_space(hdr, btp, blp, &tagp, &dup,
 386                                   &enddup, &compact, len);
 387
 388         /*
 389          * Done everything we need for a space check now.
 390          */
 391         if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
 392                 xfs_trans_brelse(tp, bp);
 393                 if (!dup)
 394                         return XFS_ERROR(ENOSPC);
 395                 return 0;
 396         }
 397
 398         /*
 399          * If we don't have space for the new entry & leaf ...
 400          */
 401         if (!dup) {
 402                 /* Don't have a space reservation: return no-space.  */
 403                 if (args->total == 0)
 404                         return XFS_ERROR(ENOSPC);
 405                 /*
 406                  * Convert to the next larger format.
 407                  * Then add the new entry in that format.
 408                  */
 409                 error = xfs_dir2_block_to_leaf(args, bp);
 410                 if (error)
 411                         return error;
 412                 return xfs_dir2_leaf_addname(args);
 413         }
 414
 415         needlog = needscan = 0;
 416
 417         /*
 418          * If need to compact the leaf entries, do it now.
 419          */
 420         if (compact) {
 421                 xfs_dir2_block_compact(tp, bp, hdr, btp, blp, &needlog,
 422                                       &lfloghigh, &lfloglow);
 423                 /* recalculate blp post-compaction */
 424                 blp = xfs_dir2_block_leaf_p(btp);
 425         } else if (btp->stale) {
 426                 /*
 427                  * Set leaf logging boundaries to impossible state.
 428                  * For the no-stale case they're set explicitly.
 429                  */
 430                 lfloglow = be32_to_cpu(btp->count);
 431                 lfloghigh = -1;
 432         }
 433
 434         /*
 435          * Find the slot that's first lower than our hash value, -1 if none.
 436          */
 437         for (low = 0, high = be32_to_cpu(btp->count) - 1; low <= high; ) {
 438                 mid = (low + high) >> 1;
 439                 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
 440                         break;
 441                 if (hash < args->hashval)
 442                         low = mid + 1;
 443                 else
 444                         high = mid - 1;
 445         }
 446         while (mid >= 0 && be32_to_cpu(blp[mid].hashval) >= args->hashval) {
 447                 mid--;
 448         }
 449         /*
 450          * No stale entries, will use enddup space to hold new leaf.
 451          */
 452         if (!btp->stale) {
 453                 /*
 454                  * Mark the space needed for the new leaf entry, now in use.
 455                  */
 456                 xfs_dir2_data_use_free(tp, bp, enddup,
 457                         (xfs_dir2_data_aoff_t)
 458                         ((char *)enddup - (char *)hdr + be16_to_cpu(enddup->length) -
 459                          sizeof(*blp)),
 460                         (xfs_dir2_data_aoff_t)sizeof(*blp),
 461                         &needlog, &needscan);
 462                 /*
 463                  * Update the tail (entry count).
 464                  */
 465                 be32_add_cpu(&btp->count, 1);
 466                 /*
 467                  * If we now need to rebuild the bestfree map, do so.
 468                  * This needs to happen before the next call to use_free.
 469                  */
 470                 if (needscan) {
 471                         xfs_dir2_data_freescan(mp, hdr, &needlog);
 472                         needscan = 0;
 473                 }
 474                 /*
 475                  * Adjust pointer to the first leaf entry, we're about to move
 476                  * the table up one to open up space for the new leaf entry.
 477                  * Then adjust our index to match.
 478                  */
 479                 blp--;
 480                 mid++;
 481                 if (mid)
 482                         memmove(blp, &blp[1], mid * sizeof(*blp));
 483                 lfloglow = 0;
 484                 lfloghigh = mid;
 485         }
 486         /*
 487          * Use a stale leaf for our new entry.
 488          */
 489         else {
 490                 for (lowstale = mid;
 491                      lowstale >= 0 &&
 492                         blp[lowstale].address !=
 493                         cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
 494                      lowstale--)
 495                         continue;
 496                 for (highstale = mid + 1;
 497                      highstale < be32_to_cpu(btp->count) &&
 498                         blp[highstale].address !=
 499                         cpu_to_be32(XFS_DIR2_NULL_DATAPTR) &&
 500                         (lowstale < 0 || mid - lowstale > highstale - mid);
 501                      highstale++)
 502                         continue;
 503                 /*
 504                  * Move entries toward the low-numbered stale entry.
 505                  */
 506                 if (lowstale >= 0 &&
 507                     (highstale == be32_to_cpu(btp->count) ||
 508                      mid - lowstale <= highstale - mid)) {
 509                         if (mid - lowstale)
 510                                 memmove(&blp[lowstale], &blp[lowstale + 1],
 511                                         (mid - lowstale) * sizeof(*blp));
 512                         lfloglow = MIN(lowstale, lfloglow);
 513                         lfloghigh = MAX(mid, lfloghigh);
 514                 }
 515                 /*
 516                  * Move entries toward the high-numbered stale entry.
 517                  */
 518                 else {
 519                         ASSERT(highstale < be32_to_cpu(btp->count));
 520                         mid++;
 521                         if (highstale - mid)
 522                                 memmove(&blp[mid + 1], &blp[mid],
 523                                         (highstale - mid) * sizeof(*blp));
 524                         lfloglow = MIN(mid, lfloglow);
 525                         lfloghigh = MAX(highstale, lfloghigh);
 526                 }
 527                 be32_add_cpu(&btp->stale, -1);
 528         }
 529         /*
 530          * Point to the new data entry.
 531          */
 532         dep = (xfs_dir2_data_entry_t *)dup;
 533         /*
 534          * Fill in the leaf entry.
 535          */
 536         blp[mid].hashval = cpu_to_be32(args->hashval);
 537         blp[mid].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
 538                                 (char *)dep - (char *)hdr));
 539         xfs_dir2_block_log_leaf(tp, bp, lfloglow, lfloghigh);
 540         /*
 541          * Mark space for the data entry used.
 542          */
 543         xfs_dir2_data_use_free(tp, bp, dup,
 544                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
 545                 (xfs_dir2_data_aoff_t)len, &needlog, &needscan);
 546         /*
 547          * Create the new data entry.
 548          */
 549         dep->inumber = cpu_to_be64(args->inumber);
 550         dep->namelen = args->namelen;
 551         memcpy(dep->name, args->name, args->namelen);
 552         tagp = xfs_dir2_data_entry_tag_p(dep);
 553         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
 554         /*
 555          * Clean up the bestfree array and log the header, tail, and entry.
 556          */
 557         if (needscan)
 558                 xfs_dir2_data_freescan(mp, hdr, &needlog);
 559         if (needlog)
 560                 xfs_dir2_data_log_header(tp, bp);
 561         xfs_dir2_block_log_tail(tp, bp);
 562         xfs_dir2_data_log_entry(tp, bp, dep);
 563         xfs_dir3_data_check(dp, bp);
 564         return 0;
 565 }
 566
 567 /*
 568  * Readdir for block directories.
 569  */
 570 int                                             /* error */
 571 xfs_dir2_block_getdents(
 572         xfs_inode_t             *dp,            /* incore inode */
 573         struct dir_context      *ctx)
 574 {
 575         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 576         struct xfs_buf          *bp;            /* buffer for block */
 577         xfs_dir2_block_tail_t   *btp;           /* block tail */
 578         xfs_dir2_data_entry_t   *dep;           /* block data entry */
 579         xfs_dir2_data_unused_t  *dup;           /* block unused entry */
 580         char                    *endptr;        /* end of the data entries */
 581         int                     error;          /* error return value */
 582         xfs_mount_t             *mp;            /* filesystem mount point */
 583         char                    *ptr;           /* current data entry */
 584         int                     wantoff;        /* starting block offset */
 585         xfs_off_t               cook;
 586
 587         mp = dp->i_mount;
 588         /*
 589          * If the block number in the offset is out of range, we're done.
 590          */
 591         if (xfs_dir2_dataptr_to_db(mp, ctx->pos) > mp->m_dirdatablk)
 592                 return 0;
 593
 594         error = xfs_dir3_block_read(NULL, dp, &bp);
 595         if (error)
 596                 return error;
 597
 598         /*
 599          * Extract the byte offset we start at from the seek pointer.
 600          * We'll skip entries before this.
 601          */
 602         wantoff = xfs_dir2_dataptr_to_off(mp, ctx->pos);
 603         hdr = bp->b_addr;
 604         xfs_dir3_data_check(dp, bp);
 605         /*
 606          * Set up values for the loop.
 607          */
 608         btp = xfs_dir2_block_tail_p(mp, hdr);
 609         ptr = (char *)xfs_dir3_data_entry_p(hdr);
 610         endptr = (char *)xfs_dir2_block_leaf_p(btp);
 611
 612         /*
 613          * Loop over the data portion of the block.
 614          * Each object is a real entry (dep) or an unused one (dup).
 615          */
 616         while (ptr < endptr) {
 617                 dup = (xfs_dir2_data_unused_t *)ptr;
 618                 /*
 619                  * Unused, skip it.
 620                  */
 621                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 622                         ptr += be16_to_cpu(dup->length);
 623                         continue;
 624                 }
 625
 626                 dep = (xfs_dir2_data_entry_t *)ptr;
 627
 628                 /*
 629                  * Bump pointer for the next iteration.
 630                  */
 631                 ptr += xfs_dir2_data_entsize(dep->namelen);
 632                 /*
 633                  * The entry is before the desired starting point, skip it.
 634                  */
 635                 if ((char *)dep - (char *)hdr < wantoff)
 636                         continue;
 637
 638                 cook = xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk,
 639                                             (char *)dep - (char *)hdr);
 640
 641                 ctx->pos = cook & 0x7fffffff;
 642                 /*
 643                  * If it didn't fit, set the final offset to here & return.
 644                  */
 645                 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
 646                             be64_to_cpu(dep->inumber), DT_UNKNOWN)) {
 647                         xfs_trans_brelse(NULL, bp);
 648                         return 0;
 649                 }
 650         }
 651
 652         /*
 653          * Reached the end of the block.
 654          * Set the offset to a non-existent block 1 and return.
 655          */
 656         ctx->pos = xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk + 1, 0) &
 657                         0x7fffffff;
 658         xfs_trans_brelse(NULL, bp);
 659         return 0;
 660 }
 661
 662 /*
 663  * Log leaf entries from the block.
 664  */
 665 static void
 666 xfs_dir2_block_log_leaf(
 667         xfs_trans_t             *tp,            /* transaction structure */
 668         struct xfs_buf          *bp,            /* block buffer */
 669         int                     first,          /* index of first logged leaf */
 670         int                     last)           /* index of last logged leaf */
 671 {
 672         xfs_dir2_data_hdr_t     *hdr = bp->b_addr;
 673         xfs_dir2_leaf_entry_t   *blp;
 674         xfs_dir2_block_tail_t   *btp;
 675
 676         btp = xfs_dir2_block_tail_p(tp->t_mountp, hdr);
 677         blp = xfs_dir2_block_leaf_p(btp);
 678         xfs_trans_log_buf(tp, bp, (uint)((char *)&blp[first] - (char *)hdr),
 679                 (uint)((char *)&blp[last + 1] - (char *)hdr - 1));
 680 }
 681
 682 /*
 683  * Log the block tail.
 684  */
 685 static void
 686 xfs_dir2_block_log_tail(
 687         xfs_trans_t             *tp,            /* transaction structure */
 688         struct xfs_buf          *bp)            /* block buffer */
 689 {
 690         xfs_dir2_data_hdr_t     *hdr = bp->b_addr;
 691         xfs_dir2_block_tail_t   *btp;
 692
 693         btp = xfs_dir2_block_tail_p(tp->t_mountp, hdr);
 694         xfs_trans_log_buf(tp, bp, (uint)((char *)btp - (char *)hdr),
 695                 (uint)((char *)(btp + 1) - (char *)hdr - 1));
 696 }
 697
 698 /*
 699  * Look up an entry in the block.  This is the external routine,
 700  * xfs_dir2_block_lookup_int does the real work.
 701  */
 702 int                                             /* error */
 703 xfs_dir2_block_lookup(
 704         xfs_da_args_t           *args)          /* dir lookup arguments */
 705 {
 706         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 707         xfs_dir2_leaf_entry_t   *blp;           /* block leaf entries */
 708         struct xfs_buf          *bp;            /* block buffer */
 709         xfs_dir2_block_tail_t   *btp;           /* block tail */
 710         xfs_dir2_data_entry_t   *dep;           /* block data entry */
 711         xfs_inode_t             *dp;            /* incore inode */
 712         int                     ent;            /* entry index */
 713         int                     error;          /* error return value */
 714         xfs_mount_t             *mp;            /* filesystem mount point */
 715
 716         trace_xfs_dir2_block_lookup(args);
 717
 718         /*
 719          * Get the buffer, look up the entry.
 720          * If not found (ENOENT) then return, have no buffer.
 721          */
 722         if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent)))
 723                 return error;
 724         dp = args->dp;
 725         mp = dp->i_mount;
 726         hdr = bp->b_addr;
 727         xfs_dir3_data_check(dp, bp);
 728         btp = xfs_dir2_block_tail_p(mp, hdr);
 729         blp = xfs_dir2_block_leaf_p(btp);
 730         /*
 731          * Get the offset from the leaf entry, to point to the data.
 732          */
 733         dep = (xfs_dir2_data_entry_t *)((char *)hdr +
 734                 xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
 735         /*
 736          * Fill in inode number, CI name if appropriate, release the block.
 737          */
 738         args->inumber = be64_to_cpu(dep->inumber);
 739         error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
 740         xfs_trans_brelse(args->trans, bp);
 741         return XFS_ERROR(error);
 742 }
 743
 744 /*
 745  * Internal block lookup routine.
 746  */
 747 static int                                      /* error */
 748 xfs_dir2_block_lookup_int(
 749         xfs_da_args_t           *args,          /* dir lookup arguments */
 750         struct xfs_buf          **bpp,          /* returned block buffer */
 751         int                     *entno)         /* returned entry number */
 752 {
 753         xfs_dir2_dataptr_t      addr;           /* data entry address */
 754         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 755         xfs_dir2_leaf_entry_t   *blp;           /* block leaf entries */
 756         struct xfs_buf          *bp;            /* block buffer */
 757         xfs_dir2_block_tail_t   *btp;           /* block tail */
 758         xfs_dir2_data_entry_t   *dep;           /* block data entry */
 759         xfs_inode_t             *dp;            /* incore inode */
 760         int                     error;          /* error return value */
 761         xfs_dahash_t            hash;           /* found hash value */
 762         int                     high;           /* binary search high index */
 763         int                     low;            /* binary search low index */
 764         int                     mid;            /* binary search current idx */
 765         xfs_mount_t             *mp;            /* filesystem mount point */
 766         xfs_trans_t             *tp;            /* transaction pointer */
 767         enum xfs_dacmp          cmp;            /* comparison result */
 768
 769         dp = args->dp;
 770         tp = args->trans;
 771         mp = dp->i_mount;
 772
 773         error = xfs_dir3_block_read(tp, dp, &bp);
 774         if (error)
 775                 return error;
 776
 777         hdr = bp->b_addr;
 778         xfs_dir3_data_check(dp, bp);
 779         btp = xfs_dir2_block_tail_p(mp, hdr);
 780         blp = xfs_dir2_block_leaf_p(btp);
 781         /*
 782          * Loop doing a binary search for our hash value.
 783          * Find our entry, ENOENT if it's not there.
 784          */
 785         for (low = 0, high = be32_to_cpu(btp->count) - 1; ; ) {
 786                 ASSERT(low <= high);
 787                 mid = (low + high) >> 1;
 788                 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
 789                         break;
 790                 if (hash < args->hashval)
 791                         low = mid + 1;
 792                 else
 793                         high = mid - 1;
 794                 if (low > high) {
 795                         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
 796                         xfs_trans_brelse(tp, bp);
 797                         return XFS_ERROR(ENOENT);
 798                 }
 799         }
 800         /*
 801          * Back up to the first one with the right hash value.
 802          */
 803         while (mid > 0 && be32_to_cpu(blp[mid - 1].hashval) == args->hashval) {
 804                 mid--;
 805         }
 806         /*
 807          * Now loop forward through all the entries with the
 808          * right hash value looking for our name.
 809          */
 810         do {
 811                 if ((addr = be32_to_cpu(blp[mid].address)) == XFS_DIR2_NULL_DATAPTR)
 812                         continue;
 813                 /*
 814                  * Get pointer to the entry from the leaf.
 815                  */
 816                 dep = (xfs_dir2_data_entry_t *)
 817                         ((char *)hdr + xfs_dir2_dataptr_to_off(mp, addr));
 818                 /*
 819                  * Compare name and if it's an exact match, return the index
 820                  * and buffer. If it's the first case-insensitive match, store
 821                  * the index and buffer and continue looking for an exact match.
 822                  */
 823                 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
 824                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
 825                         args->cmpresult = cmp;
 826                         *bpp = bp;
 827                         *entno = mid;
 828                         if (cmp == XFS_CMP_EXACT)
 829                                 return 0;
 830                 }
 831         } while (++mid < be32_to_cpu(btp->count) &&
 832                         be32_to_cpu(blp[mid].hashval) == hash);
 833
 834         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
 835         /*
 836          * Here, we can only be doing a lookup (not a rename or replace).
 837          * If a case-insensitive match was found earlier, return success.
 838          */
 839         if (args->cmpresult == XFS_CMP_CASE)
 840                 return 0;
 841         /*
 842          * No match, release the buffer and return ENOENT.
 843          */
 844         xfs_trans_brelse(tp, bp);
 845         return XFS_ERROR(ENOENT);
 846 }
 847
 848 /*
 849  * Remove an entry from a block format directory.
 850  * If that makes the block small enough to fit in shortform, transform it.
 851  */
 852 int                                             /* error */
 853 xfs_dir2_block_removename(
 854         xfs_da_args_t           *args)          /* directory operation args */
 855 {
 856         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 857         xfs_dir2_leaf_entry_t   *blp;           /* block leaf pointer */
 858         struct xfs_buf          *bp;            /* block buffer */
 859         xfs_dir2_block_tail_t   *btp;           /* block tail */
 860         xfs_dir2_data_entry_t   *dep;           /* block data entry */
 861         xfs_inode_t             *dp;            /* incore inode */
 862         int                     ent;            /* block leaf entry index */
 863         int                     error;          /* error return value */
 864         xfs_mount_t             *mp;            /* filesystem mount point */
 865         int                     needlog;        /* need to log block header */
 866         int                     needscan;       /* need to fixup bestfree */
 867         xfs_dir2_sf_hdr_t       sfh;            /* shortform header */
 868         int                     size;           /* shortform size */
 869         xfs_trans_t             *tp;            /* transaction pointer */
 870
 871         trace_xfs_dir2_block_removename(args);
 872
 873         /*
 874          * Look up the entry in the block.  Gets the buffer and entry index.
 875          * It will always be there, the vnodeops level does a lookup first.
 876          */
 877         if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
 878                 return error;
 879         }
 880         dp = args->dp;
 881         tp = args->trans;
 882         mp = dp->i_mount;
 883         hdr = bp->b_addr;
 884         btp = xfs_dir2_block_tail_p(mp, hdr);
 885         blp = xfs_dir2_block_leaf_p(btp);
 886         /*
 887          * Point to the data entry using the leaf entry.
 888          */
 889         dep = (xfs_dir2_data_entry_t *)
 890               ((char *)hdr + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
 891         /*
 892          * Mark the data entry's space free.
 893          */
 894         needlog = needscan = 0;
 895         xfs_dir2_data_make_free(tp, bp,
 896                 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
 897                 xfs_dir2_data_entsize(dep->namelen), &needlog, &needscan);
 898         /*
 899          * Fix up the block tail.
 900          */
 901         be32_add_cpu(&btp->stale, 1);
 902         xfs_dir2_block_log_tail(tp, bp);
 903         /*
 904          * Remove the leaf entry by marking it stale.
 905          */
 906         blp[ent].address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
 907         xfs_dir2_block_log_leaf(tp, bp, ent, ent);
 908         /*
 909          * Fix up bestfree, log the header if necessary.
 910          */
 911         if (needscan)
 912                 xfs_dir2_data_freescan(mp, hdr, &needlog);
 913         if (needlog)
 914                 xfs_dir2_data_log_header(tp, bp);
 915         xfs_dir3_data_check(dp, bp);
 916         /*
 917          * See if the size as a shortform is good enough.
 918          */
 919         size = xfs_dir2_block_sfsize(dp, hdr, &sfh);
 920         if (size > XFS_IFORK_DSIZE(dp))
 921                 return 0;
 922
 923         /*
 924          * If it works, do the conversion.
 925          */
 926         return xfs_dir2_block_to_sf(args, bp, size, &sfh);
 927 }
 928
 929 /*
 930  * Replace an entry in a V2 block directory.
 931  * Change the inode number to the new value.
 932  */
 933 int                                             /* error */
 934 xfs_dir2_block_replace(
 935         xfs_da_args_t           *args)          /* directory operation args */
 936 {
 937         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 938         xfs_dir2_leaf_entry_t   *blp;           /* block leaf entries */
 939         struct xfs_buf          *bp;            /* block buffer */
 940         xfs_dir2_block_tail_t   *btp;           /* block tail */
 941         xfs_dir2_data_entry_t   *dep;           /* block data entry */
 942         xfs_inode_t             *dp;            /* incore inode */
 943         int                     ent;            /* leaf entry index */
 944         int                     error;          /* error return value */
 945         xfs_mount_t             *mp;            /* filesystem mount point */
 946
 947         trace_xfs_dir2_block_replace(args);
 948
 949         /*
 950          * Lookup the entry in the directory.  Get buffer and entry index.
 951          * This will always succeed since the caller has already done a lookup.
 952          */
 953         if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
 954                 return error;
 955         }
 956         dp = args->dp;
 957         mp = dp->i_mount;
 958         hdr = bp->b_addr;
 959         btp = xfs_dir2_block_tail_p(mp, hdr);
 960         blp = xfs_dir2_block_leaf_p(btp);
 961         /*
 962          * Point to the data entry we need to change.
 963          */
 964         dep = (xfs_dir2_data_entry_t *)
 965               ((char *)hdr + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
 966         ASSERT(be64_to_cpu(dep->inumber) != args->inumber);
 967         /*
 968          * Change the inode number to the new value.
 969          */
 970         dep->inumber = cpu_to_be64(args->inumber);
 971         xfs_dir2_data_log_entry(args->trans, bp, dep);
 972         xfs_dir3_data_check(dp, bp);
 973         return 0;
 974 }
 975
 976 /*
 977  * Qsort comparison routine for the block leaf entries.
 978  */
 979 static int                                      /* sort order */
 980 xfs_dir2_block_sort(
 981         const void                      *a,     /* first leaf entry */
 982         const void                      *b)     /* second leaf entry */
 983 {
 984         const xfs_dir2_leaf_entry_t     *la;    /* first leaf entry */
 985         const xfs_dir2_leaf_entry_t     *lb;    /* second leaf entry */
 986
 987         la = a;
 988         lb = b;
 989         return be32_to_cpu(la->hashval) < be32_to_cpu(lb->hashval) ? -1 :
 990                 (be32_to_cpu(la->hashval) > be32_to_cpu(lb->hashval) ? 1 : 0);
 991 }
 992
 993 /*
 994  * Convert a V2 leaf directory to a V2 block directory if possible.
 995  */
 996 int                                             /* error */
 997 xfs_dir2_leaf_to_block(
 998         xfs_da_args_t           *args,          /* operation arguments */
 999         struct xfs_buf          *lbp,           /* leaf buffer */
1000         struct xfs_buf          *dbp)           /* data buffer */
1001 {
1002         __be16                  *bestsp;        /* leaf bests table */
1003         xfs_dir2_data_hdr_t     *hdr;           /* block header */
1004         xfs_dir2_block_tail_t   *btp;           /* block tail */
1005         xfs_inode_t             *dp;            /* incore directory inode */
1006         xfs_dir2_data_unused_t  *dup;           /* unused data entry */
1007         int                     error;          /* error return value */
1008         int                     from;           /* leaf from index */
1009         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1010         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1011         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1012         xfs_mount_t             *mp;            /* file system mount point */
1013         int                     needlog;        /* need to log data header */
1014         int                     needscan;       /* need to scan for bestfree */
1015         xfs_dir2_sf_hdr_t       sfh;            /* shortform header */
1016         int                     size;           /* bytes used */
1017         __be16                  *tagp;          /* end of entry (tag) */
1018         int                     to;             /* block/leaf to index */
1019         xfs_trans_t             *tp;            /* transaction pointer */
1020         struct xfs_dir2_leaf_entry *ents;
1021         struct xfs_dir3_icleaf_hdr leafhdr;
1022
1023         trace_xfs_dir2_leaf_to_block(args);
1024
1025         dp = args->dp;
1026         tp = args->trans;
1027         mp = dp->i_mount;
1028         leaf = lbp->b_addr;
1029         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
1030         ents = xfs_dir3_leaf_ents_p(leaf);
1031         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1032
1033         ASSERT(leafhdr.magic == XFS_DIR2_LEAF1_MAGIC ||
1034                leafhdr.magic == XFS_DIR3_LEAF1_MAGIC);
1035         /*
1036          * If there are data blocks other than the first one, take this
1037          * opportunity to remove trailing empty data blocks that may have
1038          * been left behind during no-space-reservation operations.
1039          * These will show up in the leaf bests table.
1040          */
1041         while (dp->i_d.di_size > mp->m_dirblksize) {
1042                 int hdrsz;
1043
1044                 hdrsz = xfs_dir3_data_hdr_size(xfs_sb_version_hascrc(&mp->m_sb));
1045                 bestsp = xfs_dir2_leaf_bests_p(ltp);
1046                 if (be16_to_cpu(bestsp[be32_to_cpu(ltp->bestcount) - 1]) ==
1047                                             mp->m_dirblksize - hdrsz) {
1048                         if ((error =
1049                             xfs_dir2_leaf_trim_data(args, lbp,
1050                                     (xfs_dir2_db_t)(be32_to_cpu(ltp->bestcount) - 1))))
1051                                 return error;
1052                 } else
1053                         return 0;
1054         }
1055         /*
1056          * Read the data block if we don't already have it, give up if it fails.
1057          */
1058         if (!dbp) {
1059                 error = xfs_dir3_data_read(tp, dp, mp->m_dirdatablk, -1, &dbp);
1060                 if (error)
1061                         return error;
1062         }
1063         hdr = dbp->b_addr;
1064         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
1065                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
1066
1067         /*
1068          * Size of the "leaf" area in the block.
1069          */
1070         size = (uint)sizeof(xfs_dir2_block_tail_t) +
1071                (uint)sizeof(*lep) * (leafhdr.count - leafhdr.stale);
1072         /*
1073          * Look at the last data entry.
1074          */
1075         tagp = (__be16 *)((char *)hdr + mp->m_dirblksize) - 1;
1076         dup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
1077         /*
1078          * If it's not free or is too short we can't do it.
1079          */
1080         if (be16_to_cpu(dup->freetag) != XFS_DIR2_DATA_FREE_TAG ||
1081             be16_to_cpu(dup->length) < size)
1082                 return 0;
1083
1084         /*
1085          * Start converting it to block form.
1086          */
1087         xfs_dir3_block_init(mp, tp, dbp, dp);
1088
1089         needlog = 1;
1090         needscan = 0;
1091         /*
1092          * Use up the space at the end of the block (blp/btp).
1093          */
1094         xfs_dir2_data_use_free(tp, dbp, dup, mp->m_dirblksize - size, size,
1095                 &needlog, &needscan);
1096         /*
1097          * Initialize the block tail.
1098          */
1099         btp = xfs_dir2_block_tail_p(mp, hdr);
1100         btp->count = cpu_to_be32(leafhdr.count - leafhdr.stale);
1101         btp->stale = 0;
1102         xfs_dir2_block_log_tail(tp, dbp);
1103         /*
1104          * Initialize the block leaf area.  We compact out stale entries.
1105          */
1106         lep = xfs_dir2_block_leaf_p(btp);
1107         for (from = to = 0; from < leafhdr.count; from++) {
1108                 if (ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
1109                         continue;
1110                 lep[to++] = ents[from];
1111         }
1112         ASSERT(to == be32_to_cpu(btp->count));
1113         xfs_dir2_block_log_leaf(tp, dbp, 0, be32_to_cpu(btp->count) - 1);
1114         /*
1115          * Scan the bestfree if we need it and log the data block header.
1116          */
1117         if (needscan)
1118                 xfs_dir2_data_freescan(mp, hdr, &needlog);
1119         if (needlog)
1120                 xfs_dir2_data_log_header(tp, dbp);
1121         /*
1122          * Pitch the old leaf block.
1123          */
1124         error = xfs_da_shrink_inode(args, mp->m_dirleafblk, lbp);
1125         if (error)
1126                 return error;
1127
1128         /*
1129          * Now see if the resulting block can be shrunken to shortform.
1130          */
1131         size = xfs_dir2_block_sfsize(dp, hdr, &sfh);
1132         if (size > XFS_IFORK_DSIZE(dp))
1133                 return 0;
1134
1135         return xfs_dir2_block_to_sf(args, dbp, size, &sfh);
1136 }
1137
1138 /*
1139  * Convert the shortform directory to block form.
1140  */
1141 int                                             /* error */
1142 xfs_dir2_sf_to_block(
1143         xfs_da_args_t           *args)          /* operation arguments */
1144 {
1145         xfs_dir2_db_t           blkno;          /* dir-relative block # (0) */
1146         xfs_dir2_data_hdr_t     *hdr;           /* block header */
1147         xfs_dir2_leaf_entry_t   *blp;           /* block leaf entries */
1148         struct xfs_buf          *bp;            /* block buffer */
1149         xfs_dir2_block_tail_t   *btp;           /* block tail pointer */
1150         xfs_dir2_data_entry_t   *dep;           /* data entry pointer */
1151         xfs_inode_t             *dp;            /* incore directory inode */
1152         int                     dummy;          /* trash */
1153         xfs_dir2_data_unused_t  *dup;           /* unused entry pointer */
1154         int                     endoffset;      /* end of data objects */
1155         int                     error;          /* error return value */
1156         int                     i;              /* index */
1157         xfs_mount_t             *mp;            /* filesystem mount point */
1158         int                     needlog;        /* need to log block header */
1159         int                     needscan;       /* need to scan block freespc */
1160         int                     newoffset;      /* offset from current entry */
1161         int                     offset;         /* target block offset */
1162         xfs_dir2_sf_entry_t     *sfep;          /* sf entry pointer */
1163         xfs_dir2_sf_hdr_t       *oldsfp;        /* old shortform header  */
1164         xfs_dir2_sf_hdr_t       *sfp;           /* shortform header  */
1165         __be16                  *tagp;          /* end of data entry */
1166         xfs_trans_t             *tp;            /* transaction pointer */
1167         struct xfs_name         name;
1168         struct xfs_ifork        *ifp;
1169
1170         trace_xfs_dir2_sf_to_block(args);
1171
1172         dp = args->dp;
1173         tp = args->trans;
1174         mp = dp->i_mount;
1175         ifp = XFS_IFORK_PTR(dp, XFS_DATA_FORK);
1176         ASSERT(ifp->if_flags & XFS_IFINLINE);
1177         /*
1178          * Bomb out if the shortform directory is way too short.
1179          */
1180         if (dp->i_d.di_size < offsetof(xfs_dir2_sf_hdr_t, parent)) {
1181                 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1182                 return XFS_ERROR(EIO);
1183         }
1184
1185         oldsfp = (xfs_dir2_sf_hdr_t *)ifp->if_u1.if_data;
1186
1187         ASSERT(ifp->if_bytes == dp->i_d.di_size);
1188         ASSERT(ifp->if_u1.if_data != NULL);
1189         ASSERT(dp->i_d.di_size >= xfs_dir2_sf_hdr_size(oldsfp->i8count));
1190         ASSERT(dp->i_d.di_nextents == 0);
1191
1192         /*
1193          * Copy the directory into a temporary buffer.
1194          * Then pitch the incore inode data so we can make extents.
1195          */
1196         sfp = kmem_alloc(ifp->if_bytes, KM_SLEEP);
1197         memcpy(sfp, oldsfp, ifp->if_bytes);
1198
1199         xfs_idata_realloc(dp, -ifp->if_bytes, XFS_DATA_FORK);
1200         xfs_bmap_local_to_extents_empty(dp, XFS_DATA_FORK);
1201         dp->i_d.di_size = 0;
1202
1203         /*
1204          * Add block 0 to the inode.
1205          */
1206         error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE, &blkno);
1207         if (error) {
1208                 kmem_free(sfp);
1209                 return error;
1210         }
1211         /*
1212          * Initialize the data block, then convert it to block format.
1213          */
1214         error = xfs_dir3_data_init(args, blkno, &bp);
1215         if (error) {
1216                 kmem_free(sfp);
1217                 return error;
1218         }
1219         xfs_dir3_block_init(mp, tp, bp, dp);
1220         hdr = bp->b_addr;
1221
1222         /*
1223          * Compute size of block "tail" area.
1224          */
1225         i = (uint)sizeof(*btp) +
1226             (sfp->count + 2) * (uint)sizeof(xfs_dir2_leaf_entry_t);
1227         /*
1228          * The whole thing is initialized to free by the init routine.
1229          * Say we're using the leaf and tail area.
1230          */
1231         dup = xfs_dir3_data_unused_p(hdr);
1232         needlog = needscan = 0;
1233         xfs_dir2_data_use_free(tp, bp, dup, mp->m_dirblksize - i, i, &needlog,
1234                 &needscan);
1235         ASSERT(needscan == 0);
1236         /*
1237          * Fill in the tail.
1238          */
1239         btp = xfs_dir2_block_tail_p(mp, hdr);
1240         btp->count = cpu_to_be32(sfp->count + 2);       /* ., .. */
1241         btp->stale = 0;
1242         blp = xfs_dir2_block_leaf_p(btp);
1243         endoffset = (uint)((char *)blp - (char *)hdr);
1244         /*
1245          * Remove the freespace, we'll manage it.
1246          */
1247         xfs_dir2_data_use_free(tp, bp, dup,
1248                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
1249                 be16_to_cpu(dup->length), &needlog, &needscan);
1250         /*
1251          * Create entry for .
1252          */
1253         dep = xfs_dir3_data_dot_entry_p(hdr);
1254         dep->inumber = cpu_to_be64(dp->i_ino);
1255         dep->namelen = 1;
1256         dep->name[0] = '.';
1257         tagp = xfs_dir2_data_entry_tag_p(dep);
1258         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1259         xfs_dir2_data_log_entry(tp, bp, dep);
1260         blp[0].hashval = cpu_to_be32(xfs_dir_hash_dot);
1261         blp[0].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
1262                                 (char *)dep - (char *)hdr));
1263         /*
1264          * Create entry for ..
1265          */
1266         dep = xfs_dir3_data_dotdot_entry_p(hdr);
1267         dep->inumber = cpu_to_be64(xfs_dir2_sf_get_parent_ino(sfp));
1268         dep->namelen = 2;
1269         dep->name[0] = dep->name[1] = '.';
1270         tagp = xfs_dir2_data_entry_tag_p(dep);
1271         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1272         xfs_dir2_data_log_entry(tp, bp, dep);
1273         blp[1].hashval = cpu_to_be32(xfs_dir_hash_dotdot);
1274         blp[1].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
1275                                 (char *)dep - (char *)hdr));
1276         offset = xfs_dir3_data_first_offset(hdr);
1277         /*
1278          * Loop over existing entries, stuff them in.
1279          */
1280         i = 0;
1281         if (!sfp->count)
1282                 sfep = NULL;
1283         else
1284                 sfep = xfs_dir2_sf_firstentry(sfp);
1285         /*
1286          * Need to preserve the existing offset values in the sf directory.
1287          * Insert holes (unused entries) where necessary.
1288          */
1289         while (offset < endoffset) {
1290                 /*
1291                  * sfep is null when we reach the end of the list.
1292                  */
1293                 if (sfep == NULL)
1294                         newoffset = endoffset;
1295                 else
1296                         newoffset = xfs_dir2_sf_get_offset(sfep);
1297                 /*
1298                  * There should be a hole here, make one.
1299                  */
1300                 if (offset < newoffset) {
1301                         dup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
1302                         dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1303                         dup->length = cpu_to_be16(newoffset - offset);
1304                         *xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16(
1305                                 ((char *)dup - (char *)hdr));
1306                         xfs_dir2_data_log_unused(tp, bp, dup);
1307                         xfs_dir2_data_freeinsert(hdr, dup, &dummy);
1308                         offset += be16_to_cpu(dup->length);
1309                         continue;
1310                 }
1311                 /*
1312                  * Copy a real entry.
1313                  */
1314                 dep = (xfs_dir2_data_entry_t *)((char *)hdr + newoffset);
1315                 dep->inumber = cpu_to_be64(xfs_dir2_sfe_get_ino(sfp, sfep));
1316                 dep->namelen = sfep->namelen;
1317                 memcpy(dep->name, sfep->name, dep->namelen);
1318                 tagp = xfs_dir2_data_entry_tag_p(dep);
1319                 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1320                 xfs_dir2_data_log_entry(tp, bp, dep);
1321                 name.name = sfep->name;
1322                 name.len = sfep->namelen;
1323                 blp[2 + i].hashval = cpu_to_be32(mp->m_dirnameops->
1324                                                         hashname(&name));
1325                 blp[2 + i].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
1326                                                  (char *)dep - (char *)hdr));
1327                 offset = (int)((char *)(tagp + 1) - (char *)hdr);
1328                 if (++i == sfp->count)
1329                         sfep = NULL;
1330                 else
1331                         sfep = xfs_dir2_sf_nextentry(sfp, sfep);
1332         }
1333         /* Done with the temporary buffer */
1334         kmem_free(sfp);
1335         /*
1336          * Sort the leaf entries by hash value.
1337          */
1338         xfs_sort(blp, be32_to_cpu(btp->count), sizeof(*blp), xfs_dir2_block_sort);
1339         /*
1340          * Log the leaf entry area and tail.
1341          * Already logged the header in data_init, ignore needlog.
1342          */
1343         ASSERT(needscan == 0);
1344         xfs_dir2_block_log_leaf(tp, bp, 0, be32_to_cpu(btp->count) - 1);
1345         xfs_dir2_block_log_tail(tp, bp);
1346         xfs_dir3_data_check(dp, bp);
1347         return 0;
1348 }